In [1]:
import pandas as pd
import numpy as np
import seaborn as sns
import scipy.stats as stats
import matplotlib.pyplot as plt
%matplotlib inline
import networkx as nx
import plotly
from plotly.offline import download_plotlyjs, init_notebook_mode, iplot
import plotly.graph_objs as go
import plotly.figure_factory as ff
import plotly.express as px
px.set_mapbox_access_token(open("./mapbox_token").read())
# pip install plotly_express==0.4.0
from bs4 import BeautifulSoup
from warnings import filterwarnings
filterwarnings('ignore')
from datetime import datetime
# Data Source
# https://www.kaggle.com/kimjihoo/coronavirusdataset?select=SeoulFloating.csv
from sklearn.model_selection import train_test_split
# Algorithms
from sklearn import linear_model
from sklearn.linear_model import LogisticRegression
from sklearn.ensemble import RandomForestClassifier
from sklearn.linear_model import Perceptron
from sklearn.linear_model import SGDClassifier
from sklearn.tree import DecisionTreeClassifier
from sklearn.neighbors import KNeighborsClassifier
from sklearn.svm import SVC, LinearSVC
from sklearn.naive_bayes import GaussianNB
# Matrix
from sklearn.metrics import precision_score, recall_score
from sklearn.model_selection import cross_val_predict
from sklearn.metrics import confusion_matrix
# Diagram
from sklearn.tree import export_graphviz
import pydotplus
from sklearn.externals.six import StringIO
from IPython.display import Image
Enable Table Of Content
pip install jupyter_contrib_nbextensions jupyter contrib nbextension install --user jupyter nbextension enable spellchecker/main jupyter nbextension enable codefolding/main
Exploratory Data Analysis¶
Case¶
Overview of the cases in Korea¶
In [2]:
caseData = pd.read_csv('covid/case.csv')
caseDataForMap = caseData.copy()
caseDataForMap = caseDataForMap[caseDataForMap[['latitude', 'longitude']] != '-']
cols = ['latitude', 'longitude']
caseData[cols] = caseData[cols].apply(pd.to_numeric, errors='coerce', axis=1)
display(caseData.head())
fig = px.scatter_mapbox(
caseData[caseData.latitude != '-'],
text="<br>City: " + caseData["city"] +" <br>Province: " + caseData["province"],
lat="latitude",
lon="longitude",
color="confirmed",
size="confirmed",
color_continuous_scale=px.colors.sequential.Burg,
size_max=100,
mapbox_style='dark',
zoom=6,
title="Overview of the cases in Korea")
fig.show()
Check the relationship between group and solo cases¶
In [3]:
caseData = pd.read_csv('covid/case.csv')
caseDataForMap = caseData.copy()
caseDataSorted = caseDataForMap.sort_values(by=['confirmed'], ascending=False)
display(caseDataSorted.head())
sortedValues = caseDataForMap.groupby(['province','group']).sum().sort_values(by=['confirmed'], ascending=False).reset_index()
sortedValues = sortedValues[['province','confirmed','group']]
display(sortedValues.head())
sortedValues['logConfirmed'] = np.log(sortedValues['confirmed'])
chart = sns.factorplot(x='confirmed', y='province', hue='group',data=sortedValues, kind='bar', height=6.27, aspect=17.7/8.27)
chart.set_xticklabels(rotation=90)
Out[3]:
Source of infected cases sort by source and province¶
In [4]:
#Used log, because daegu has high amount of infection which highly skewed the data
caseData['logConfirmed'] = np.log(caseData['confirmed'])
locations = caseData.pivot("province", "infection_case", "logConfirmed")
display(locations.head())
f, ax = plt.subplots(figsize=(20, 5))
sns.heatmap(locations, cmap="RdPu", annot=False, fmt="d", linewidths=.5, ax=ax).set_title('Numbers of cases sort by source and province')
Out[4]:
Risk Level of each province¶
In [5]:
caseData = pd.read_csv('covid/case.csv')
caseDataForMap = caseData.copy()
caseDataForMap['logConfirmed'] = np.log(caseDataForMap['confirmed'])
a4_dims = (14.7, 8.27)
fig, ax = plt.subplots(figsize=a4_dims)
chart = sns.boxplot(x='logConfirmed', y='province',data=caseDataForMap, ax=ax)
In [6]:
caseDataForMap = caseData.copy()
sortedValues = caseDataForMap.groupby(['province']).sum().sort_values(by=['confirmed'], ascending=False).reset_index()
sortedValues['logConfirmed'] = np.log(sortedValues['confirmed'])
sortedValues = sortedValues[['province','confirmed','logConfirmed']]
display(sortedValues.head())
In [7]:
print('Mean: ', sortedValues.mean().values)
print('Standard Derivation: ', sortedValues.std().values)
This amount of COVID-19 cases has a mean of 670 and a standard deviation of 1610.
Hence we sort the provinces into 4 different types of level:
- Very Risk = 4 (Above 2000 cases)
- High Risk = 3 (1000 - 2000 cases)
- Medium Risk = 2 (100 - 999 cases)
- Low Risk = 1 ( Below 99 cases)
In [8]:
def calculate_risk_level(sortedValues) :
if sortedValues['confirmed'] >= 2000:
return 4
elif sortedValues['confirmed'] >= 1000 and sortedValues['confirmed'] < 2000:
return 3
elif sortedValues['confirmed'] >= 100 and sortedValues['confirmed'] < 900:
return 2
else:
return 1
sortedValues['Risk Level'] = sortedValues.apply(calculate_risk_level,axis=1);
sortedValues = sortedValues[['province','Risk Level']]
display(sortedValues.head())
Patient Info¶
Numbers Of Cases Per Day¶
In [9]:
patientData = pd.read_csv('covid/patientinfo.csv')
display(patientData.head())
groupPatientData = patientData.groupby('confirmed_date').size().reset_index()
groupPatientData.columns = ['confirmed_date', 'count']
fig = px.line(groupPatientData, x="confirmed_date", y="count", title='Numbers Of Covid Cases Per Day')
fig.show()
Average Survivor Treatment Day (Recovery Speed), By Age and Gender¶
In [10]:
confinedDaysAnalysis = patientData.copy()
confinedDaysAnalysis = confinedDaysAnalysis[confinedDaysAnalysis['age'] != '100s'] #Just a single data which is highly skewed
confinedDaysAnalysis = confinedDaysAnalysis[confinedDaysAnalysis['released_date'].notnull()]
cols = ['released_date', 'confirmed_date']
confinedDaysAnalysis[cols] = confinedDaysAnalysis[cols].apply(pd.to_datetime, errors='coerce', axis=1)
confinedDaysAnalysis['Total Treatment Days'] = (confinedDaysAnalysis['released_date'] - confinedDaysAnalysis['confirmed_date']).dt.days
groupedData = confinedDaysAnalysis.groupby(['sex', 'age'])['Total Treatment Days'].mean().unstack().stack().reset_index()
groupedData.columns = ["sex", "age", "Average Treatment Days"]
dataForHeatmap = groupedData.pivot("sex", "age", "Average Treatment Days")
display(dataForHeatmap.head())
f, ax = plt.subplots(figsize=(15, 5))
sns.heatmap(dataForHeatmap, cmap="RdPu", annot=False, fmt="d", linewidths=.5, ax=ax).set_title('Average Survivor Treatment Day, Sort by Age & Gender')
Out[10]:
In [11]:
confinedDaysAnalysis['age'] = confinedDaysAnalysis['age'].dropna()
age = {'0s':0, '10s':1, '20s':2, '30s':3, '40s':4, '50s':5, '60s':6, '70s':7, '80s':8, '90s':9}
confinedDaysAnalysis['age'] = confinedDaysAnalysis['age'].map(age)
confinedDaysAnalysis = confinedDaysAnalysis[confinedDaysAnalysis['age'].notna()]
fig = px.scatter(confinedDaysAnalysis, x="age", y="Total Treatment Days", color="sex", trendline="ols", title="Does age group affect total treatment days?")
fig.show()
Average Survivor Treatment Day (Recovery Speed) By City & Age¶
In [12]:
groupedData = confinedDaysAnalysis.groupby(['province', 'age'])['Total Treatment Days'].mean().unstack().stack().reset_index()
groupedData.columns = ["province", "age", "Average Treatment Days"]
groupedData['Average Treatment Days'] = np.ceil(groupedData['Average Treatment Days'].apply(pd.to_numeric, errors='coerce'))
dataForHeatmap = groupedData.pivot("province", "age", "Average Treatment Days")
display(dataForHeatmap.head())
f, ax = plt.subplots(figsize=(20, 5))
sns.heatmap(dataForHeatmap, cmap="RdPu", annot=False, fmt="d", linewidths=.5, ax=ax).set_title('Average Survivor Treatment Day, Sort by City & Age')
Out[12]:
Average Deceased Treatment Day (Fatality Speed) By Gender and Age¶
In [13]:
confinedDaysAnalysis = patientData.copy()
confinedDaysAnalysis = confinedDaysAnalysis[confinedDaysAnalysis['age'] != '100s'] #Just a single data which is highly skewed
confinedDaysAnalysis = confinedDaysAnalysis[confinedDaysAnalysis['deceased_date'].notnull()]
cols = ['deceased_date', 'confirmed_date']
confinedDaysAnalysis[cols] = confinedDaysAnalysis[cols].apply(pd.to_datetime, errors='coerce', axis=1)
confinedDaysAnalysis['Total Treatment Days'] = (confinedDaysAnalysis['deceased_date'] - confinedDaysAnalysis['confirmed_date']).dt.days
display(confinedDaysAnalysis.head())
groupedData = confinedDaysAnalysis.groupby(['sex', 'age'])['Total Treatment Days'].mean().unstack().stack().reset_index()
groupedData.columns = ["sex", "age", "Average Treatment Days"]
dataForHeatmap = groupedData.pivot("sex", "age", "Average Treatment Days")
display(dataForHeatmap.head())
f, ax = plt.subplots(figsize=(15, 5))
sns.heatmap(dataForHeatmap, cmap="RdPu", annot=False, fmt="d", linewidths=.5, ax=ax).set_title('Average Deceased Treatment Day, Sort by Age & Gender')
Out[13]:
Average Deceased Treatment Day (Fatality Speed) By City and Age¶
In [14]:
groupedData = confinedDaysAnalysis.groupby(['province', 'age'])['Total Treatment Days'].mean().unstack().stack().reset_index()
groupedData.columns = ["province", "age", "Average Treatment Days"]
groupedData['Average Treatment Days'] = np.ceil(groupedData['Average Treatment Days'].apply(pd.to_numeric, errors='coerce'))
dataForHeatmap = groupedData.pivot("province", "age", "Average Treatment Days")
display(dataForHeatmap.head())
f, ax = plt.subplots(figsize=(20, 5))
sns.heatmap(dataForHeatmap, cmap="RdPu", annot=False, fmt="d", linewidths=.5, ax=ax).set_title('Average Deceased Treatment Day, Sort by City & Age')
Out[14]:
Number of Infection, Survivor, Deceased Per City by percentage¶
In [15]:
survivorCountAnalysis = patientData.copy()
survivorCountAnalysis['survive'] = survivorCountAnalysis['released_date'].notnull()
survivorCountAnalysis['deceased'] = survivorCountAnalysis['deceased_date'].notnull()
survivorCountAnalysis['under treatment'] = survivorCountAnalysis['deceased_date'].isnull() & survivorCountAnalysis['released_date'].isnull()
provinceStats = survivorCountAnalysis.groupby(['province']).sum()
provinceStatsClean = provinceStats[['survive', 'deceased', 'under treatment']]
provinceStatsClean['survive %'] = np.round(provinceStatsClean['survive'] / (provinceStatsClean['survive'] + provinceStatsClean['deceased'] + provinceStatsClean['under treatment']) * 100, 2)
provinceStatsClean['deceased %'] = np.round(provinceStatsClean['deceased'] / (provinceStatsClean['survive'] + provinceStatsClean['deceased'] + provinceStatsClean['under treatment']) * 100, 2)
provinceStatsClean['under treatment %'] = np.round(provinceStatsClean['under treatment'] / (provinceStatsClean['survive'] + provinceStatsClean['deceased'] + provinceStatsClean['under treatment']) * 100, 2)
provinceStatsAbsolute = provinceStatsClean[['survive', 'deceased', 'under treatment']]
provinceStatsPercentage = provinceStatsClean[['survive %', 'deceased %', 'under treatment %']]
display(provinceStatsAbsolute.head())
display(provinceStatsPercentage.head())
In [16]:
newTable = provinceStatsPercentage.reset_index()
display(newTable.head())
newTable = pd.melt(newTable, id_vars="province", var_name="stats", value_name="rate")
chart = sns.factorplot(x='rate', y='province', hue='stats', data=newTable, kind='bar', height=6.27, aspect=16.7/8.27)
chart.set_xticklabels(rotation=90)
Out[16]:
Number of Infection, Survivor, Deceased Sort By Gender¶
In [17]:
survivorCountAnalysis = patientData.copy()
survivorCountAnalysis['survive'] = survivorCountAnalysis['released_date'].notnull()
survivorCountAnalysis['deceased'] = survivorCountAnalysis['deceased_date'].notnull()
survivorCountAnalysis['under treatment'] = survivorCountAnalysis['deceased_date'].isnull() & survivorCountAnalysis['released_date'].isnull()
provinceStats = survivorCountAnalysis.groupby(['sex']).sum()
provinceStatsClean = provinceStats[['survive', 'deceased', 'under treatment']]
provinceStatsClean['survive %'] = np.round(provinceStatsClean['survive'] / (provinceStatsClean['survive'] + provinceStatsClean['deceased'] + provinceStatsClean['under treatment']) * 100, 2)
provinceStatsClean['deceased %'] = np.round(provinceStatsClean['deceased'] / (provinceStatsClean['survive'] + provinceStatsClean['deceased'] + provinceStatsClean['under treatment']) * 100, 2)
provinceStatsClean['under treatment %'] = np.round(provinceStatsClean['under treatment'] / (provinceStatsClean['survive'] + provinceStatsClean['deceased'] + provinceStatsClean['under treatment']) * 100, 2)
provinceStatsAbsolute = provinceStatsClean[['survive', 'deceased', 'under treatment']]
provinceStatsPercentage = provinceStatsClean[['survive %', 'deceased %', 'under treatment %']]
display(provinceStatsAbsolute)
display(provinceStatsPercentage)
In [18]:
newTable = provinceStatsPercentage.reset_index()
newTable = pd.melt(newTable, id_vars="sex", var_name="stats", value_name="rate")
display(newTable.head())
chart = sns.factorplot(x='rate', y='sex', hue='stats', data=newTable, kind='bar', height=6.27, aspect=17.7/8.27)
chart.set_xticklabels(rotation=90)
Out[18]:
Network Diagram¶
In [19]:
networkData = patientData.copy()
networkData = networkData[networkData['infected_by'].notnull()]
networkData = networkData[['patient_id','sex','age','province','city','infection_case','infected_by','state']]
display(networkData.head())
In [20]:
A = list(networkData["infected_by"].unique())
B = list(networkData["patient_id"].unique())
node_list = set(A+B)
# Create Graph
G = nx.Graph()
for i in node_list:
G.add_node(i)
# G.nodes()
for i,j in networkData.iterrows():
G.add_edges_from([(j["infected_by"],j["patient_id"])])
pos = nx.spring_layout(G, k=0.5, iterations=50)
for n, p in pos.items():
G.nodes[n]['pos'] = p
edge_trace = go.Scatter(
x=[],
y=[],
line=dict(width=0.5,color='#888'),
hoverinfo='none',
mode='lines')
for edge in G.edges():
x0, y0 = G.nodes[edge[0]]['pos']
x1, y1 = G.nodes[edge[1]]['pos']
edge_trace['x'] += tuple([x0, x1, None])
edge_trace['y'] += tuple([y0, y1, None])
node_trace = go.Scatter(
x=[],
y=[],
text=[],
mode='markers',
hoverinfo='text',
marker=dict(
showscale=True,
colorscale='RdPu_r',
reversescale=True,
color=[],
size=15,
colorbar=dict(
thickness=10,
title='Numnber of Infected Cases',
xanchor='left',
titleside='right'
),
line=dict(width=0)))
for node in G.nodes():
x, y = G.nodes[node]['pos']
node_trace['x'] += tuple([x])
node_trace['y'] += tuple([y])
for node, adjacencies in enumerate(G.adjacency()):
node_trace['marker']['color']+=tuple([len(adjacencies[1])])
node_info = str(adjacencies[0]) +' # of connections: '+ str(len(adjacencies[1]))
node_trace['text']+=tuple([node_info])
fig = go.Figure(data=[edge_trace, node_trace],
layout=go.Layout(
title='<br>Korea Covid Network Connections',
titlefont=dict(size=16),
showlegend=False,
hovermode='closest',
margin=dict(b=20,l=5,r=5,t=40),
annotations=[ dict(
text="",
showarrow=False,
xref="paper", yref="paper") ],
xaxis=dict(showgrid=False, zeroline=False, showticklabels=False),
yaxis=dict(showgrid=False, zeroline=False, showticklabels=False)))
iplot(fig)
Policy¶
In [21]:
policyData = pd.read_csv('covid/policy.csv')
policyDataCopy = policyData.copy()
policyDataCopy['end_date'] = policyDataCopy['end_date'].fillna(datetime.now().strftime('%Y-%m-%d'))
display(policyDataCopy.head())
df = []
for index, row in policyDataCopy.iterrows():
df.append(dict(Task=row['gov_policy'], Start=row['start_date'], Finish=row['end_date'], Resource=row['type']))
fig = px.timeline(df, x_start="Start", x_end="Finish", y="Task", color="Resource", title="Policy over time")
fig.show()
patientData = pd.read_csv('covid/patientinfo.csv')
groupPatientData = patientData.groupby('confirmed_date').size().reset_index()
groupPatientData.columns = ['confirmed_date', 'count']
fig = px.line(groupPatientData, x="confirmed_date", y="count", title='Numbers Of Covid Cases Per Day')
fig.show()
Time¶
In [22]:
timeData = pd.read_csv('covid/time.csv')
timeDataMelted = pd.melt(timeData, id_vars=['date'], value_vars=['test', 'negative','confirmed','released','deceased'])
display(timeDataMelted.head())
fig = px.line(timeDataMelted, x="date", y="value", color='variable', title="Overall cases over time")
fig.show()
In [23]:
timeDataMelted = pd.melt(timeData, id_vars=['date'], value_vars=['confirmed','released','deceased'])
display(timeDataMelted.head())
fig = px.line(timeDataMelted, x="date", y="value", color='variable', title="Confirmed, released & deceased over time")
fig.show()
TimeAge¶
In [24]:
timeAgeData = pd.read_csv('covid/timeage.csv')
display(timeAgeData.head())
fig = px.line(timeAgeData, x="date", y="confirmed", color='age', title="Confirmed cases of various age group over time")
fig.show()
In [25]:
timeAgeData = pd.read_csv('covid/timeage.csv')
fig = px.line(timeAgeData, x="date", y="deceased", color='age', title="Deceased cases of various age group over time")
fig.show()
Time Gender¶
In [26]:
timeGender = pd.read_csv('covid/timegender.csv')
display(timeGender.head())
fig = px.line(timeGender, x="date", y="confirmed", color='sex', title="Confirmed cases between genders over time")
fig.show()
In [27]:
fig = px.line(timeGender, x="date", y="deceased", color='sex', title="Deceased cases between genders over time")
fig.show()
Time Province¶
In [28]:
timeProvince = pd.read_csv('covid/timeprovince.csv')
display(timeProvince.head())
fig = px.line(timeProvince, x="date", y="confirmed", color='province', title="Confirmed cases of various province over time")
fig.show()
In [29]:
fig = px.line(timeProvince, x="date", y="released", color='province', title="Released cases of various province over time")
fig.show()
In [30]:
fig = px.line(timeProvince, x="date", y="deceased", color='province', title="Deceased cases of various province over time")
fig.show()
Time Series World Map Visualisation of COVID-19 cases in korea¶¶
In [31]:
timeProvince = pd.read_csv('covid/timeprovince.csv')
data = {'province': ['Seoul', 'Busan', 'Daegu', 'Incheon', 'Gwangju', 'Daejeon',
'Ulsan', 'Sejong', 'Gyeonggi-do', 'Gangwon-do',
'Chungcheongbuk-do', 'Chungcheongnam-do', 'Jeollabuk-do',
'Jeollanam-do', 'Gyeongsangbuk-do', 'Gyeongsangnam-do', 'Jeju-do'],
'longitude': [127.047325,129.066666,128.600006,126.705208,126.916664,127.385002,
129.316666,127.2822,127.143738,127.920158,
127.935905,126.979874,126.916664,126.9910,
129.263885,128.429581,126.5312],
'latitude': [37.517235,35.166668,35.866669,37.456257,35.166668,36.351002,
35.549999,36.4870,37.603405,37.342220,
36.981304,36.806702,35.166668,34.8679,
35.835354,34.855228,33.4996]}
location = pd.DataFrame(data=data)
mergedData = timeProvince.merge(location, on='province', how='left')
mergedData = mergedData.dropna()
mergedData.head()
Out[31]:
With Animation¶
In [32]:
fig = px.scatter_mapbox(
mergedData, lat="latitude", lon="longitude",
size="confirmed", size_max=100,
color="deceased", color_continuous_scale=px.colors.sequential.Burg,
hover_name="province",
mapbox_style='dark', zoom=6,
animation_frame="date", animation_group="province"
)
fig.layout.updatemenus[0].buttons[0].args[1]["frame"]["duration"] = 200
fig.layout.updatemenus[0].buttons[0].args[1]["transition"]["duration"] = 200
fig.layout.coloraxis.showscale = False
fig.layout.sliders[0].pad.t = 10
fig.layout.updatemenus[0].pad.t= 10
fig.show()
Without animation¶
In [33]:
fig = px.scatter_mapbox(
mergedData, lat="latitude", lon="longitude",
size="confirmed", size_max=100,
color="deceased", color_continuous_scale=px.colors.sequential.Burg,
hover_name="province",
mapbox_style='dark', zoom=6,
)
fig.show()
Weather¶
In [34]:
weatherData = pd.read_csv('covid/weather.csv')
weatherData = weatherData.set_index('date')
weatherData = weatherData['2020-01-01' : '2020-08-31'].reset_index()
display(weatherData.head())
weatherDataForHeatmap = weatherData.pivot("date", "province", "avg_temp")
f, ax = plt.subplots(figsize=(20, 20))
sns.heatmap(weatherDataForHeatmap, cmap="RdPu", annot=False, fmt="d", linewidths=.5, ax=ax).set_title('Weather Over Time')
Out[34]:
Region¶
In [35]:
regionData = pd.read_csv('covid/region.csv')
pd.set_option('display.max_rows', regionData.shape[0]+1)
display(regionData.head())
fig = px.scatter_mapbox(
regionData[regionData.city != 'Korea'],
text="city",
lat="latitude",
lon="longitude",
color="elderly_population_ratio",
size="nursing_home_count",
color_continuous_scale=px.colors.sequential.Burg,
size_max=100,
zoom=6,
title="Number of nusing home and elder population ratio across Korea")
fig.show()
Web Scalping: Comparsion across the world¶
In [36]:
import requests
page = requests.get("https://www.worldometers.info/coronavirus")
page.status_code
Out[36]:
In [37]:
# page.content
soup = BeautifulSoup(page.content, 'lxml')
In [38]:
# print(soup.prettify())
In [39]:
covidTable = soup.find('table', attrs={'id': 'main_table_countries_today'})
# covidTable
Getting the data from the table¶
In [40]:
rows = covidTable.find_all("tr", attrs={"style": ""})
In [41]:
covidData = []
for i,data in enumerate(rows):
if i == 0:
covidData.append(data.text.strip().split("\n")[:13])
else:
covidData.append(data.text.strip().split("\n")[:12])
# covidData
Covert to data¶
In [42]:
covidTable = pd.DataFrame(covidData[1:], columns=covidData[0][:12])
covidTable = covidTable[~covidTable['#'].isin(['World', 'Total:'])]
covidTable = covidTable.drop('#', axis =1)
covidTable.head()
Out[42]:
Worldwide Total Cases Chart¶
In [43]:
fig = px.bar(covidTable, y='TotalCases', x='Country,Other')
fig.show()
We will be using log because the data is highly skewed
In [44]:
covidTable['logTotalCases'] = np.log(covidTable['TotalCases'].str.replace(r'\D', '').astype(int))
covidTable = covidTable.sort_values(by=['logTotalCases'])
fig = px.bar(covidTable, y='logTotalCases', x='Country,Other')
fig.show()
Worldwide Total Death Chart¶
In [45]:
pd.set_option('display.max_rows', covidTable.shape[0]+1)
covidTable['TotalDeaths'].str.strip()
covidTable['TotalDeaths'].replace('', np.nan, inplace=True)
covidTable['TotalDeaths'].replace(' ', np.nan, inplace=True)
covidTable['TotalDeaths'] = covidTable['TotalDeaths'].fillna(0)
covidTable['logTotalDeath'] = np.log(covidTable['TotalDeaths'].str.replace(r'\D', '').astype(float))
covidTable = covidTable.sort_values(by=['logTotalDeath'])
fig = px.bar(covidTable, y='logTotalDeath', x='Country,Other')
fig.show()
Plotting Data¶
Summarise Key findings from Exploratory Data Analysis¶
Overall Cases in Korea¶
- On average, there are 100 COVID-19 cases in infected city.
- The numbers are highly skewed by Nam-gu, a province of Daegu. Nam-gu has about 4500 cases.
- Most of the clusters are in the province of Daegu, Gyeonggi-do and Seoul
Cases across various Korea province¶
- Majority of the province has more individual cases of COVID-19 than group cases with the exception of Incheon and Ulsan
- Seoul, Daegu, Gyeonggi-do and Gyeongsangbuk-do has the most number of individual cases among the Korea provinces, number of cases range from 150 to 200 with the exception of Daegu
- In addition, the same group of provinces has the most number of grouped cases, number of cases range from 50-100 with the exception of Daegu
Source of infection¶
- Sources of infection mainly come from Shincheonji Church, Itawon Clubs, Contact with Patients and Overseas inflow
- Most of the infection sources across various province has less than 500 cases with the exception of Daegu.
- Daegu has more than 4000 cases from Shincheonji Church and about 2000 cases from contact with patient and other cases each.
Patient Cases¶
- There is a quick spike in daily COVID-19 cases from Feb to Mar before slowing dripping down to single digit in May. There are double digit COVID-19 cases in June
Recovery Speed¶
- Age has an R-squared value of 0.04 for male and 0.03 for female against total treatment days.
- At first graze, age seems to have little to no effect on recovery speed.
- However when we used the mean recovery days for each age group, it was found that younger people recovered fast from COVID-19 as compared to older people.
- Average days to recover for young people (<=30) is around 25 days and below. Middle-aged people(30-60) took about 25-30 days to recover. Older people (>=70) took about 32-36 days to recover
- Interestingly, infected people in Daegu only conist of people from 50s to 70s. In addition, it took less than 15 days to recover. People in Jeollabuk-do take longer to recover from COVID-19 as compared to the rest of the provinces. Majority of the people took about 20-25 days to recover.
Fatality Speed¶
- The average number of treatment day a patient had before he/she passed away is shorter than the average day it took for a patient recover.
- In addition, younger patient has a lower number of treatment days as compared to older patients. This might indicate that younger patient seek treatment at the later stage of COVID-19.
- Coincidentally, the number of treatment days increased as the age increases. The trend is similar to the recovery speed among during age group.
- Daegu has the fastest fatality speed. An average patient is under treatment for less than 10 days before he/she passed away
- Gyeonggi-do has an outliner. A 30s year old patient died in last than 10 days after seeking treatment.
Percentage of patients recovered, under treatment and deceased¶
- Daegu has the highest fatality amount all the provinces. It has a fatality rate of more than 10%. In addition, Daejeon, Gyeongsangbuk-do and Ulsan are the only provinces which have fatailty.
- Male has a higher fatality rate and a lowest survival rate as compared to female
COVID-19 Tranmission Amount¶
- Majority of the cases only spread to 1 other person.
- Cases #20000000205 is a super spread who spreaded to more than 50 people.
- In addition, there are less than 10 cases where patient spread to more than 30 people or more.
Policy¶
- There is a quick spike in daily COVID-19 cases from Feb to Mar before slowing dripping down to single digit in May. There are double digit COVID-19 cases in June
- Technology related policy are introduced in Feb, especially thanks to policy like open data, people like us are able to do analysis on Korea COVID-19 cases.
- After the spike in COVID-19 cases in March, education, social, health and immigration policy are introduced rapidly. Furthermore, the alert level was upgraded to Level 4 red.
- After the introduction of the policies, the number of COVID-19 cases dropped drastically during April and May. Hence the policies are effective.
- However, it seems to be lax in control of COVID-19 cases as the COVID-19 slowly creep up to double digit in May/June. This forced the govtnment to introduced administrative policies such as closure of clubs to control the situation.
Cases Over Time Analysis¶
- As for 31st Jun 2020, 1.27 million of people has gone for testing, 1.24 million are tested negative, 12.8 thousand of confirmed cases, 11.5 thousand of released cases and 282 people has passed away.
- Majority of the people are tested negative for covid-19 and less than 1% are tested positive for the virus.
- Among those who are tested positive, almost 90% of the people has recovered and about 2% of the people has passed away
Cases Over Age & Time Analysis¶
- Age group in the 20s 40s 50s 60s has higher infection cases as compared to other age group.
- Furthermore, in the deceased section, most of the fatalities are in the age group of 60s 70s 80s,
- This indicate that the risk of death from COVID-19 increases with age.
Cases Over Gender & Time Analysis¶
- There are more female infected cases as compared to male cases.
- The rate of increases of confirmed cases are about the same for both gender over the last 6 months
- On the other hand, male are more prone to death to COVID-19 as compared to female
Cases Over Province & Time Analysis¶
- The province of Daegu, Gyeongsangbuk-do, Incheon and Seoul has the most amount of covid-19 infection cases among the 17 provinces; with Daegu accounting for more than half of the cases.
- The fatality count is relatively proportional to the infection count with the except of Seoul
- Bother Seoul and Gyeonggi-do has about 900 infected cases. However, Gyeonggi-do has about 22 fatalities and Seoul has about 7 fatalities.
Weather¶
- The weather is about the same across the provinces.
- The temperature gradually increases from around 5 degree to 25 degree for the period of Jan 2020 to June 2020
- The weather is relatively the same across all provinces, it has little to do with the reported cases in the short term
Number of nursing home vs Population Ratio¶
- We observed an interesting fact regarding the elderly population ratio and the number of nursing home in the city.
- The elderly population ratio and the number of nursing home has an inverse relationship
- Perhaps the absolute number of the elder people is smaller where elderly population ratio are high which explains why there is nursing home in places with higher elderly population ratio.
Preparing the data for modelling¶
We will create a model which will predict whether a COVID-19 patient will survive given a certain set of conditions
- Sort city & province into different risk level according to number of confirmed cases (Low, Medium, High)
- Sort age into different age level according to age (kid, adult, elderly)
- Gender into categorical group of 0 & 1
- Number of treatment day
- Status of released, deceased into different category (We will ignore cases who are under treatment)
Get the necessary columns¶
In [46]:
patientData = pd.read_csv('covid/patientinfo.csv')
patientModellingData = patientData[['sex','age','province','confirmed_date','released_date','deceased_date','state']]
## We removed isolated patient since it is not confirmed if they survived or not
patientModellingData = patientModellingData[patientModellingData['state'] != 'isolated']
nullList = ['sex','age','confirmed_date']
for item in nullList:
patientModellingData = patientModellingData[~patientModellingData[item].isnull()]
display(patientModellingData.head())
Calculate number of days of treatment before dropped date column¶
In [47]:
cols = ['released_date', 'deceased_date', 'confirmed_date']
patientModellingData[cols] = patientModellingData[cols].apply(pd.to_datetime, errors='coerce', axis=1)
def calculate_number_of_treatment_days(row):
if row["released_date"] is not pd.NaT:
treatmentDays = pd.to_numeric((row['released_date'] - row['confirmed_date']).days)
return(treatmentDays)
elif row["deceased_date"] is not pd.NaT:
treatmentDays = pd.to_numeric((row['deceased_date'] - row['confirmed_date']).days)
return(treatmentDays)
else:
return None
patientModellingData['Treatment Days'] = patientModellingData.apply(calculate_number_of_treatment_days, axis=1)
patientModellingData = patientModellingData[~patientModellingData['Treatment Days'].isnull()]
patientModellingDataTreatment = patientModellingData[['sex','age','province','state','Treatment Days']]
display(patientModellingDataTreatment.head())
Convert state, gender and gender columns to categorical data¶
In [48]:
genders = {"male": 0, "female": 1}
patientModellingDataTreatment['sex'] = patientModellingDataTreatment['sex'].map(genders)
display(patientModellingDataTreatment.head())
In [49]:
state = {"released": 0, "deceased": 1}
patientModellingDataTreatment['state'] = patientModellingDataTreatment['state'].map(state)
display(patientModellingDataTreatment.head())
In [50]:
age = {'0s':0, '10s':1, '20s':2, '30s':3, '40s':4, '50s':5, '60s':6, '70s':7, '80s':8, '90s':9}
patientModellingDataTreatment['age'] = patientModellingDataTreatment['age'].map(age)
display(patientModellingDataTreatment.head())
Get dummy for province¶
In [51]:
provinceDummy = pd.get_dummies(patientModellingDataTreatment['province'])
provinceDummy
Out[51]:
In [52]:
dataForModelling = patientModellingDataTreatment.join(provinceDummy)
dataForModelling = dataForModelling.drop('province',axis=1)
dataForModelling
Out[52]:
Correlation Matrix¶
In [53]:
plt.figure(figsize=(10,5))
sns.heatmap(dataForModelling.corr())
dataForModelling['age'] = pd.to_numeric(dataForModelling['age'])
dataForModellingForCleaning = dataForModelling.copy()
display(dataForModellingForCleaning.count())
print(np.any(np.isnan(dataForModellingForCleaning)))
print(np.all(np.isfinite(dataForModellingForCleaning)))
def clean_dataset(df):
assert isinstance(df, pd.DataFrame), "df needs to be a pd.DataFrame"
df.dropna(inplace=True)
indices_to_keep = ~df.isin([np.nan, np.inf, -np.inf]).any(1)
return df[indices_to_keep].astype(np.float64)
clean_dataset(dataForModellingForCleaning)
display(dataForModellingForCleaning.count())
print(np.any(np.isnan(dataForModellingForCleaning)))
print(np.all(np.isfinite(dataForModellingForCleaning)))
Building Machine Learning Models Part 1¶
In [54]:
x = dataForModellingForCleaning.drop("state", axis=1)
y = dataForModellingForCleaning["state"]
X_train, X_test, Y_train, Y_test = train_test_split(x, y, test_size=0.33, random_state=42)
Stochastic Gradient Descent (SGD):¶
In [55]:
sgd = linear_model.SGDClassifier(max_iter=5, tol=None)
sgd.fit(X_train, Y_train)
Y_pred = sgd.predict(X_test)
sgd.score(X_train, Y_train)
acc_sgd = round(sgd.score(X_train, Y_train) * 100, 2)
Random Forest:¶
In [56]:
random_forest = RandomForestClassifier(n_estimators=100)
random_forest.fit(X_train, Y_train)
Y_prediction = random_forest.predict(X_test)
random_forest.score(X_train, Y_train)
acc_random_forest = round(random_forest.score(X_train, Y_train) * 100, 2)
Logistic Regression:¶
In [57]:
logreg = LogisticRegression()
logreg.fit(X_train, Y_train)
Y_pred = logreg.predict(X_test)
acc_log = round(logreg.score(X_train, Y_train) * 100, 2)
Gaussian Naive Bayes:¶
In [58]:
gaussian = GaussianNB()
gaussian.fit(X_train, Y_train)
Y_pred = gaussian.predict(X_test)
acc_gaussian = round(gaussian.score(X_train, Y_train) * 100, 2)
K Nearest Neighbor:¶
In [59]:
knn = KNeighborsClassifier(n_neighbors = 3)
knn.fit(X_train, Y_train)
Y_pred = knn.predict(X_test)
acc_knn = round(knn.score(X_train, Y_train) * 100, 2)
Perceptron:¶
In [60]:
perceptron = Perceptron(max_iter=5)
perceptron.fit(X_train, Y_train)
Y_pred = perceptron.predict(X_test)
acc_perceptron = round(perceptron.score(X_train, Y_train) * 100, 2)
Linear Support Vector Machine:¶
In [61]:
linear_svc = LinearSVC()
linear_svc.fit(X_train, Y_train)
Y_pred = linear_svc.predict(X_test)
acc_linear_svc = round(linear_svc.score(X_train, Y_train) * 100, 2)
Decision Tree¶
In [62]:
decision_tree = DecisionTreeClassifier()
decision_tree.fit(X_train, Y_train)
Y_pred = decision_tree.predict(X_test)
acc_decision_tree = round(decision_tree.score(X_train, Y_train) * 100, 2)
Getting the best model¶
In [63]:
results = pd.DataFrame({
'Model': ['Support Vector Machines', 'KNN', 'Logistic Regression',
'Random Forest', 'Naive Bayes', 'Perceptron',
'Stochastic Gradient Decent',
'Decision Tree'],
'Score': [acc_linear_svc, acc_knn, acc_log,
acc_random_forest, acc_gaussian, acc_perceptron,
acc_sgd, acc_decision_tree]})
result_df = results.sort_values(by='Score', ascending=False)
result_df = result_df.set_index('Score')
result_df.head(9)
Out[63]:
Decision Tree Diagram¶
In [64]:
feature_cols = x.columns
dot_data = StringIO()
export_graphviz(decision_tree, out_file = dot_data,
feature_names = feature_cols,
filled = True, rounded = True,
special_characters = True)
graph = pydotplus.graph_from_dot_data(dot_data.getvalue())
graph.write_png('covidTree.png')
Image(graph.create_png())
# Entropy 0 == no disorder // perfect knowledge, perfect classification
Out[64]:
What is the most important feature ?¶
In [65]:
importances = pd.DataFrame({'feature':X_train.columns,'importance':np.round(random_forest.feature_importances_,3)})
importances1 = importances.sort_values('importance',ascending=False).set_index('feature')
importances1.head(15)
Out[65]:
In [66]:
barData = importances.reset_index()
fig = px.bar(barData, x='feature', y='importance')
fig.show()
Confusion Matrix with Precision & Recall¶
In [67]:
predictions = cross_val_predict(random_forest, X_train, Y_train, cv=3)
result1_train = confusion_matrix(Y_train, predictions)
display(result1_train)
p1_train = precision_score(Y_train, predictions)
r1_train = recall_score(Y_train, predictions)
print("Precision:", p1_train)
print("Recall:", r1_train)
In [68]:
predictions = cross_val_predict(random_forest, X_test, Y_test, cv=3)
result1_test = confusion_matrix(Y_test, predictions)
display(result1_test)
p1_test = precision_score(Y_test, predictions)
r1_test = recall_score(Y_test, predictions)
print("Precision:", p1_test)
print("Recall:", r1_test)
Building Machine Learning Models Part 2¶
Lets observe what happened if we drop the treatment days column Will the model be more accurate?
In [69]:
x2 = dataForModellingForCleaning.drop(["state","Treatment Days"], axis=1)
y2 = dataForModellingForCleaning["state"]
X_train, X_test, Y_train, Y_test = train_test_split(x2, y2, test_size=0.33, random_state=42)
Random Forest¶
In [70]:
random_forest = RandomForestClassifier(n_estimators=100)
random_forest.fit(X_train, Y_train)
Y_prediction = random_forest.predict(X_test)
random_forest.score(X_train, Y_train)
acc_random_forest = round(random_forest.score(X_train, Y_train) * 100, 2)
Decision Tree¶
In [71]:
decision_tree = DecisionTreeClassifier()
decision_tree.fit(X_train, Y_train)
Y_pred = decision_tree.predict(X_test)
acc_decision_tree = round(decision_tree.score(X_train, Y_train) * 100, 2)
Getting the best model¶
In [72]:
results = pd.DataFrame({
'Model': ['Random Forest','Decision Tree'],
'Score': [acc_random_forest, acc_decision_tree]})
result_df = results.sort_values(by='Score', ascending=False)
result_df = result_df.set_index('Score')
result_df.head(2)
Out[72]:
Decision Tree Diagram¶
In [73]:
feature_cols = x2.columns
dot_data = StringIO()
export_graphviz(decision_tree, out_file = dot_data,
feature_names = feature_cols,
filled = True, rounded = True,
special_characters = True)
graph = pydotplus.graph_from_dot_data(dot_data.getvalue())
graph.write_png('covidTree2.png')
Image(graph.create_png())
Out[73]:
Importance¶
In [74]:
importances = pd.DataFrame({'feature':X_train.columns,'importance':np.round(random_forest.feature_importances_,3)})
importances2 = importances.sort_values('importance',ascending=False).set_index('feature')
importances2.head(15)
Out[74]:
In [75]:
barData = importances.reset_index()
fig = px.bar(barData, x='feature', y='importance')
fig.show()
Confusion Matrix with Precision & Recall¶
In [76]:
predictions = cross_val_predict(random_forest, X_train, Y_train, cv=3)
result2_train = confusion_matrix(Y_train, predictions)
display(result2_train)
p2_train = precision_score(Y_train, predictions)
r2_train = recall_score(Y_train, predictions)
print("Precision:", p2_train)
print("Recall:", r2_train)
In [77]:
predictions = cross_val_predict(random_forest, X_test, Y_test, cv=3)
result2_test = confusion_matrix(Y_test, predictions)
display(result2_test)
p2_test = precision_score(Y_test, predictions)
r2_test = recall_score(Y_test, predictions)
print("Precision:", p2_test)
print("Recall:", r2_test)
Summary between the old and new model¶
The old model is better than the new model even though it has a lower precision rate.
The number of survivor in COVID-19 is much higher than the number of deceased.
Hence, it is much harder to predict who will pass away from the infection.
The new model is unable to predict whether a patient will pass away.
The number of false negatives outweigh the number of true negatives.
On the other hand, the old model is able to predict the deceased much accurately than the new model.
With that in mind, this also indicated that the number of days of treatment is vital.
Step to take to control the COVID-19 situation even better¶
Generally, those who passed away from the infection had lesser days of treatment. However, is it right for us to conclude that lesser amount of treatment days, the higher the fatality rate? Are the risk of dying higher during first few days/weeks of treatment? Or is there something more to it?
Abraham Wald and the Missing Bullet Holes¶
This situation is something similar to Abraham Wald and the Missing Bullet Holes. @penguinpress summed up the situation perfectly:
You don't want your planes to get shot down by enemy fighters, so you armor them. But armor makes the plane heavier, and heavier planes are less maneuverable and use more fuel. Armoring the planes too much is a problem; armoring the planes too little is a problem. Somewhere in between there's an optimum.
When American planes came back from engagements over Europe, they were covered in bullet holes. But the damage wasn't uniformly distributed across the aircraft. There were more bullet holes in the fuselage, not so many in the engines.
At first glance, it seems reasonable to focus the armor on the fuselage. However, the armor, said Wald, doesn't go where the bullet holes are. It goes where the bullet holes aren't: on the engines.
The missing bullet holes were on the missing planes. The reason planes were coming back with fewer hits to the engine is that planes that got hit in the engine weren't coming back.
If you go to the recovery room at the hospital, you'll see a lot more people with bullet holes in their legs than people with bullet holes in their chests. But that's not because people don't get shot in the chest; it's because the people who get shot in the chest don't recover.
The link to the excellent article written by @penguinpress is in the credit section below.
Survivorship Bias¶
In this case, our model only consists of people who have passed away after they have gone through treatment. We have excluded people who have died even before they have the opportunity to go through treatment!
Even after we have excluded those who have not undergo treatment before they passed away, our model indicated that the number of treatment days is the strongest factor among gender, location and age. Treatment days has a whooping percentage of 60% in terms of importance next to age which has a importance of 20%.
Improving the situation¶
The chances of survival increases as the number of treatment days increases. Hence, the faster detection rate, the earlier the treatment, the more days and opportunities to treat the infection, the better the chances of survival. The Korea government put a very strong emphasis on COVID-19 detection programme and hence they managed to contain the COVID-19 situation.
Various cities among the world were also in lockdown to prevent the rapid spread of COVID-19. The healthcare system will be overwhelmed if COVID-19 were to spread widely which will cause the death rate to escalate as more people will have lesser opportunity to undergo treatment.
The older the person, the higher the chances of infection and death. Hence it is not advisable for elderly people to roam around unless it is necessary.
In terms of locations, the majority of the cases came from Daegu and the sources of infection is from places such as church and clubs. The higher the amount of contact, the greater the chances of infection. It is advisable for the authority to close such areas until a vaccine for COVID-19 is found. During May to June, there is an increase in COVID-19 cases due to complacency.
https://www.aa.com.tr/en/asia-pacific/s-korea-sees-mass-covid-19-cases-linked-to-night-clubs/1838031
https://www.channelnewsasia.com/news/asia/south-korea-covid-19-church-backlash-13092284
Bonus: Refine Dataset for Machine Learning¶
Will the model's accuracy improve if we include patient who died without any treatment?
In [78]:
patientData = pd.read_csv('covid/patientinfo.csv')
patientModellingData = patientData[['sex','age','province','confirmed_date','released_date','deceased_date','state']]
deadPatientModellingData = patientData[patientData['state'] == 'deceased']
totalDiedPatient = deadPatientModellingData.count()
display(totalDiedPatient)
display('***********')
pd.set_option('display.max_rows', deadPatientModellingData.shape[0]+1)
totalDiedPaitentWithoutTreatment = deadPatientModellingData[deadPatientModellingData['deceased_date'].isnull() == True].count()
display(totalDiedPaitentWithoutTreatment)
Assuming patient who died without any released_date or deceased_date are those died without any treatment
There are about 12 patient who died without any treatment
In [79]:
str(12/78 * 100) + '%'
Out[79]:
This accounts for 15% of the total death
Preparing the data for modelling¶
We will simply add another line of logic to return treatment days = 0
if the patient is deceased and both released and deceased date are empty
In [80]:
patientData = pd.read_csv('covid/patientinfo.csv')
patientModellingData = patientData[['sex','age','province','confirmed_date','released_date','deceased_date','state']]
## We removed isolated patient since it is not confirmed if they survived or not
patientModellingData = patientModellingData[patientModellingData['state'] != 'isolated']
nullList = ['sex','age','confirmed_date']
for item in nullList:
patientModellingData = patientModellingData[~patientModellingData[item].isnull()]
display(patientModellingData.head())
In [81]:
cols = ['released_date', 'deceased_date', 'confirmed_date']
patientModellingData[cols] = patientModellingData[cols].apply(pd.to_datetime, errors='coerce', axis=1)
def calculate_number_of_treatment_days(row):
if row["released_date"] is not pd.NaT:
treatmentDays = pd.to_numeric((row['released_date'] - row['confirmed_date']).days)
return(treatmentDays)
elif row["deceased_date"] is not pd.NaT:
treatmentDays = pd.to_numeric((row['deceased_date'] - row['confirmed_date']).days)
return(treatmentDays)
elif row["deceased_date"] is pd.NaT and row["released_date"] is pd.NaT:
if row["state"] == 'deceased' and row["state"] != 'released':
return 0
else:
return None
else:
return None
patientModellingData['Treatment Days'] = patientModellingData.apply(calculate_number_of_treatment_days, axis=1)
patientModellingData = patientModellingData[~patientModellingData['Treatment Days'].isnull()]
patientModellingDataTreatment = patientModellingData[['sex','age','province','state','Treatment Days']]
display(patientModellingDataTreatment.head())
In [82]:
genders = {"male": 0, "female": 1}
patientModellingDataTreatment['sex'] = patientModellingDataTreatment['sex'].map(genders)
state = {"released": 0, "deceased": 1}
patientModellingDataTreatment['state'] = patientModellingDataTreatment['state'].map(state)
age = {'0s':0, '10s':1, '20s':2, '30s':3, '40s':4, '50s':5, '60s':6, '70s':7, '80s':8, '90s':9}
patientModellingDataTreatment['age'] = patientModellingDataTreatment['age'].map(age)
display(patientModellingDataTreatment.head())
In [83]:
dataForModelling = patientModellingDataTreatment.join(provinceDummy)
dataForModelling = dataForModelling.drop('province',axis=1)
dataForModelling
Out[83]:
Building Machine Learning Models Part 3¶
In [84]:
clean_dataset(dataForModelling)
display(dataForModelling.count())
print(np.any(np.isnan(dataForModelling)))
print(np.all(np.isfinite(dataForModelling)))
x3 = dataForModelling.drop(["state"], axis=1)
y3 = dataForModelling["state"]
X_train, X_test, Y_train, Y_test = train_test_split(x3, y3, test_size=0.33, random_state=42)
## Random_forest
random_forest = RandomForestClassifier(n_estimators=100)
random_forest.fit(X_train, Y_train)
Y_prediction = random_forest.predict(X_test)
random_forest.score(X_train, Y_train)
acc_random_forest = round(random_forest.score(X_train, Y_train) * 100, 2)
## Decision Tree
decision_tree = DecisionTreeClassifier()
decision_tree.fit(X_train, Y_train)
Y_pred = decision_tree.predict(X_test)
acc_decision_tree = round(decision_tree.score(X_train, Y_train) * 100, 2)
results = pd.DataFrame({
'Model': ['Random Forest','Decision Tree'],
'Score': [acc_random_forest, acc_decision_tree]})
result_df = results.sort_values(by='Score', ascending=False)
result_df = result_df.set_index('Score')
result_df.head(2)
Out[84]:
Importances¶
In [85]:
importances = pd.DataFrame({'feature':X_train.columns,'importance':np.round(random_forest.feature_importances_,3)})
importances3 = importances.sort_values('importance',ascending=False).set_index('feature')
importances3.head(15)
Out[85]:
Confusion Matrix with Precision & Recall¶
In [86]:
predictions = cross_val_predict(random_forest, X_train, Y_train, cv=3)
result3_train = confusion_matrix(Y_train, predictions)
display(result3_train)
p3_train = precision_score(Y_train, predictions)
r3_train = recall_score(Y_train, predictions)
print("Precision:", p3_train)
print("Recall:", r3_train)
In [87]:
predictions = cross_val_predict(random_forest, X_test, Y_test, cv=3)
result3_test = confusion_matrix(Y_test, predictions)
display(result3_test)
p3_test = precision_score(Y_test, predictions)
r3_test = recall_score(Y_test, predictions)
print("Precision:", p3_test)
print("Recall:", r3_test)
Final Summary¶
In [88]:
print('Model with treatment day :')
print('importance:', importances1.head(3))
display(result1_train)
display(result1_test)
print('******************************')
print('Model without treatment day: ')
print('importance:', importances2.head(3))
display(result2_train)
display(result2_test)
print('******************************')
print('Model with treatment day and those who died without treatment:')
print('importance:', importances3.head(3))
display(result3_train)
display(result3_test)
print('******************************')
print('Model with treatment day :')
print("Train Data Precision:", p1_train)
print("Train Data Recall:", r1_train)
print("Test Data Precision:", p1_test)
print("Test Data Recall:", r1_test)
print('******************************')
print('Model without treatment day: ')
print("Train Data Precision:", p2_train)
print("Train Data Recall:", r2_train)
print("Test Data Precision:", p2_test)
print("Test Data Recall:", r2_test)
print('******************************')
print('Model with treatment day and those who died without treatment:')
print("Train Data Precision:", p3_train)
print("Train Data Recall:", r3_train)
print("Test Data Precision:", p3_test)
print("Test Data Recall:", r3_test)
Credit¶
Abraham Wald and the Missing Bullet Holes:
https://medium.com/@penguinpress/an-excerpt-from-how-not-to-be-wrong-by-jordan-ellenberg-664e708cfc3d\n
Decision Tree Diagram:
Ze Yu's sample project
In [ ]: